[Tmrg] Proposal to increase TCP initial CWND

lachlan.andrew at gmail.com (Lachlan Andrew) Mon, 19 July 2010 11:03 UTC

From: lachlan.andrew at gmail.com (Lachlan Andrew)
Date: Mon, 19 Jul 2010 21:03:24 +1000
Subject: [Tmrg] Proposal to increase TCP initial CWND
In-Reply-To: <AANLkTimFEGfPJo9rjTw6oNPmwuHkEFBmAX1Ps2yYsUWh@mail.gmail.com>
References: <AANLkTil937lyUzRvUtdqd2qdl9RN7AZ-Mo_cT-dtmqXz@mail.gmail.com> <923833DA-FC31-4D30-96A5-48D4D072DBD3@cisco.com> <AANLkTimFEGfPJo9rjTw6oNPmwuHkEFBmAX1Ps2yYsUWh@mail.gmail.com>
Message-ID: <AANLkTin_3NYVm_AQA9lboe5WIcMX9VJOTBthNJb13HWE@mail.gmail.com>

Greetings Jerry,

On 19 July 2010 18:12, Jerry Chu <hkchu at google.com> wrote:
> I'd like to point out Google's services are not restricted to just
> broadband users. Also from our experiements
> some of the greatest latency improvments of a larger IW seem to have
> come from users behind slow or highly
> multiplexed links.

Can you distinguish between slow links and highly multiplexed ones?
If so, that is great!  That is exactly the sort of technique I was
hoping would emerge from this discussion.  If you can already do that,
can you tell us how?

Otherwise, if the biggest gains actually come from links which are
highly multiplexed, that may be a danger sign.  That is what you would
expect from a scheme which redistributes capacity rather than
improving efficiency.

As Matt pointed out earlier, using IW=10 gives benefits from being
able to do fast-retransmit more often, which also helps on small BDP
links.  One way to separate that effect is to try an IW of 5.  That is
large enough that fast-retransmit can be used all of the time, but
will have less other impact than setting IW=10.  Did you take any
measurements like that?


Lachlan Andrew? Centre for Advanced Internet Architectures (CAIA)
Swinburne University of Technology, Melbourne, Australia
Ph +61 3 9214 4837